Commit e8a0464c authored by David S. Miller's avatar David S. Miller

netdev: Allocate multiple queues for TX.

alloc_netdev_mq() now allocates an array of netdev_queue
structures for TX, based upon the queue_count argument.

Furthermore, all accesses to the TX queues are now vectored
through the netdev_get_tx_queue() and netdev_for_each_tx_queue()
interfaces.  This makes it easy to grep the tree for all
things that want to get to a TX queue of a net device.

Problem spots which are not really multiqueue aware yet, and
only work with one queue, can easily be spotted by grepping
for all netdev_get_tx_queue() calls that pass in a zero index.
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
parent 070825b3
......@@ -5042,7 +5042,9 @@ static int bond_check_params(struct bond_params *params)
static struct lock_class_key bonding_netdev_xmit_lock_key;
static void bond_set_lockdep_class_one(struct netdev_queue *txq)
static void bond_set_lockdep_class_one(struct net_device *dev,
struct netdev_queue *txq,
void *_unused)
{
lockdep_set_class(&txq->_xmit_lock,
&bonding_netdev_xmit_lock_key);
......@@ -5050,7 +5052,7 @@ static void bond_set_lockdep_class_one(struct netdev_queue *txq)
static void bond_set_lockdep_class(struct net_device *dev)
{
bond_set_lockdep_class_one(&dev->tx_queue);
netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
}
/* Create a new bond based on the specified name and bonding parameters.
......
......@@ -124,14 +124,16 @@ static LIST_HEAD(bpq_devices);
*/
static struct lock_class_key bpq_netdev_xmit_lock_key;
static void bpq_set_lockdep_class_one(struct netdev_queue *txq)
static void bpq_set_lockdep_class_one(struct net_device *dev,
struct netdev_queue *txq,
void *_unused)
{
lockdep_set_class(&txq->_xmit_lock, &bpq_netdev_xmit_lock_key);
}
static void bpq_set_lockdep_class(struct net_device *dev)
{
bpq_set_lockdep_class_one(&dev->tx_queue);
netdev_for_each_tx_queue(dev, bpq_set_lockdep_class_one, NULL);
}
/* ------------------------------------------------------------------------ */
......
......@@ -229,14 +229,20 @@ module_param(numifbs, int, 0);
MODULE_PARM_DESC(numifbs, "Number of ifb devices");
/*
* dev_ifb->tx_queue.lock is usually taken after dev->rx_queue.lock,
* dev_ifb's TX queue lock is usually taken after dev->rx_queue.lock,
* reversely to e.g. qdisc_lock_tree(). It should be safe until
* ifb doesn't take dev->tx_queue.lock with dev_ifb->rx_queue.lock.
* ifb doesn't take dev's TX queue lock with dev_ifb->rx_queue.lock.
* But lockdep should know that ifb has different locks from dev.
*/
static struct lock_class_key ifb_tx_queue_lock_key;
static struct lock_class_key ifb_rx_queue_lock_key;
static void set_tx_lockdep_key(struct net_device *dev,
struct netdev_queue *txq,
void *_unused)
{
lockdep_set_class(&txq->lock, &ifb_tx_queue_lock_key);
}
static int __init ifb_init_one(int index)
{
......@@ -258,7 +264,7 @@ static int __init ifb_init_one(int index)
if (err < 0)
goto err;
lockdep_set_class(&dev_ifb->tx_queue.lock, &ifb_tx_queue_lock_key);
netdev_for_each_tx_queue(dev_ifb, set_tx_lockdep_key, NULL);
lockdep_set_class(&dev_ifb->rx_queue.lock, &ifb_rx_queue_lock_key);
return 0;
......
......@@ -285,7 +285,9 @@ static struct lock_class_key macvlan_netdev_xmit_lock_key;
#define MACVLAN_STATE_MASK \
((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
static void macvlan_set_lockdep_class_one(struct netdev_queue *txq)
static void macvlan_set_lockdep_class_one(struct net_device *dev,
struct netdev_queue *txq,
void *_unused)
{
lockdep_set_class(&txq->_xmit_lock,
&macvlan_netdev_xmit_lock_key);
......@@ -293,7 +295,7 @@ static void macvlan_set_lockdep_class_one(struct netdev_queue *txq)
static void macvlan_set_lockdep_class(struct net_device *dev)
{
macvlan_set_lockdep_class_one(&dev->tx_queue);
netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
}
static int macvlan_init(struct net_device *dev)
......
......@@ -3102,7 +3102,9 @@ static void prism2_clear_set_tim_queue(local_info_t *local)
*/
static struct lock_class_key hostap_netdev_xmit_lock_key;
static void prism2_set_lockdep_class_one(struct netdev_queue *txq)
static void prism2_set_lockdep_class_one(struct net_device *dev,
struct netdev_queue *txq,
void *_unused)
{
lockdep_set_class(&txq->_xmit_lock,
&hostap_netdev_xmit_lock_key);
......@@ -3110,7 +3112,7 @@ static void prism2_set_lockdep_class_one(struct netdev_queue *txq)
static void prism2_set_lockdep_class(struct net_device *dev)
{
prism2_set_lockdep_class_one(&dev->tx_queue);
netdev_for_each_tx_queue(dev, prism2_set_lockdep_class_one, NULL);
}
static struct net_device *
......
......@@ -463,7 +463,7 @@ struct netdev_queue {
struct Qdisc *qdisc_sleeping;
struct list_head qdisc_list;
struct netdev_queue *next_sched;
};
} ____cacheline_aligned_in_smp;
/*
* The DEVICE structure.
......@@ -641,7 +641,9 @@ struct net_device
unsigned char broadcast[MAX_ADDR_LEN]; /* hw bcast add */
struct netdev_queue rx_queue;
struct netdev_queue tx_queue ____cacheline_aligned_in_smp;
struct netdev_queue *_tx ____cacheline_aligned_in_smp;
unsigned int num_tx_queues;
unsigned long tx_queue_len; /* Max frames per queue allowed */
/*
......@@ -764,6 +766,25 @@ struct net_device
#define NETDEV_ALIGN 32
#define NETDEV_ALIGN_CONST (NETDEV_ALIGN - 1)
static inline
struct netdev_queue *netdev_get_tx_queue(const struct net_device *dev,
unsigned int index)
{
return &dev->_tx[index];
}
static inline void netdev_for_each_tx_queue(struct net_device *dev,
void (*f)(struct net_device *,
struct netdev_queue *,
void *),
void *arg)
{
unsigned int i;
for (i = 0; i < dev->num_tx_queues; i++)
f(dev, &dev->_tx[i], arg);
}
/*
* Net namespace inlines
*/
......@@ -977,7 +998,7 @@ static inline void netif_schedule_queue(struct netdev_queue *txq)
static inline void netif_schedule(struct net_device *dev)
{
netif_schedule_queue(&dev->tx_queue);
netif_schedule_queue(netdev_get_tx_queue(dev, 0));
}
/**
......@@ -993,7 +1014,7 @@ static inline void netif_tx_start_queue(struct netdev_queue *dev_queue)
static inline void netif_start_queue(struct net_device *dev)
{
netif_tx_start_queue(&dev->tx_queue);
netif_tx_start_queue(netdev_get_tx_queue(dev, 0));
}
/**
......@@ -1017,7 +1038,7 @@ static inline void netif_tx_wake_queue(struct netdev_queue *dev_queue)
static inline void netif_wake_queue(struct net_device *dev)
{
netif_tx_wake_queue(&dev->tx_queue);
netif_tx_wake_queue(netdev_get_tx_queue(dev, 0));
}
/**
......@@ -1034,7 +1055,7 @@ static inline void netif_tx_stop_queue(struct netdev_queue *dev_queue)
static inline void netif_stop_queue(struct net_device *dev)
{
netif_tx_stop_queue(&dev->tx_queue);
netif_tx_stop_queue(netdev_get_tx_queue(dev, 0));
}
/**
......@@ -1050,7 +1071,7 @@ static inline int netif_tx_queue_stopped(const struct netdev_queue *dev_queue)
static inline int netif_queue_stopped(const struct net_device *dev)
{
return netif_tx_queue_stopped(&dev->tx_queue);
return netif_tx_queue_stopped(netdev_get_tx_queue(dev, 0));
}
/**
......@@ -1134,7 +1155,7 @@ static inline void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
#endif
if (test_and_clear_bit(__QUEUE_STATE_XOFF,
&dev->egress_subqueue[queue_index].state))
__netif_schedule(&dev->tx_queue);
__netif_schedule(netdev_get_tx_queue(dev, 0));
}
/**
......@@ -1430,18 +1451,19 @@ static inline void __netif_tx_lock(struct netdev_queue *txq, int cpu)
static inline void netif_tx_lock(struct net_device *dev)
{
__netif_tx_lock(&dev->tx_queue, smp_processor_id());
}
int cpu = smp_processor_id();
unsigned int i;
static inline void __netif_tx_lock_bh(struct netdev_queue *txq)
{
spin_lock_bh(&txq->_xmit_lock);
txq->xmit_lock_owner = smp_processor_id();
for (i = 0; i < dev->num_tx_queues; i++) {
struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
__netif_tx_lock(txq, cpu);
}
}
static inline void netif_tx_lock_bh(struct net_device *dev)
{
__netif_tx_lock_bh(&dev->tx_queue);
local_bh_disable();
netif_tx_lock(dev);
}
static inline int __netif_tx_trylock(struct netdev_queue *txq)
......@@ -1454,7 +1476,7 @@ static inline int __netif_tx_trylock(struct netdev_queue *txq)
static inline int netif_tx_trylock(struct net_device *dev)
{
return __netif_tx_trylock(&dev->tx_queue);
return __netif_tx_trylock(netdev_get_tx_queue(dev, 0));
}
static inline void __netif_tx_unlock(struct netdev_queue *txq)
......@@ -1465,18 +1487,19 @@ static inline void __netif_tx_unlock(struct netdev_queue *txq)
static inline void netif_tx_unlock(struct net_device *dev)
{
__netif_tx_unlock(&dev->tx_queue);
}
unsigned int i;
for (i = 0; i < dev->num_tx_queues; i++) {
struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
__netif_tx_unlock(txq);
}
static inline void __netif_tx_unlock_bh(struct netdev_queue *txq)
{
txq->xmit_lock_owner = -1;
spin_unlock_bh(&txq->_xmit_lock);
}
static inline void netif_tx_unlock_bh(struct net_device *dev)
{
__netif_tx_unlock_bh(&dev->tx_queue);
netif_tx_unlock(dev);
local_bh_enable();
}
#define HARD_TX_LOCK(dev, txq, cpu) { \
......
......@@ -230,32 +230,47 @@ extern void tcf_destroy_chain(struct tcf_proto **fl);
/* Reset all TX qdiscs of a device. */
static inline void qdisc_reset_all_tx(struct net_device *dev)
{
qdisc_reset(dev->tx_queue.qdisc);
unsigned int i;
for (i = 0; i < dev->num_tx_queues; i++)
qdisc_reset(netdev_get_tx_queue(dev, i)->qdisc);
}
/* Are all TX queues of the device empty? */
static inline bool qdisc_all_tx_empty(const struct net_device *dev)
{
const struct netdev_queue *txq = &dev->tx_queue;
const struct Qdisc *q = txq->qdisc;
unsigned int i;
for (i = 0; i < dev->num_tx_queues; i++) {
struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
const struct Qdisc *q = txq->qdisc;
return (q->q.qlen == 0);
if (q->q.qlen)
return false;
}
return true;
}
/* Are any of the TX qdiscs changing? */
static inline bool qdisc_tx_changing(struct net_device *dev)
{
struct netdev_queue *txq = &dev->tx_queue;
return (txq->qdisc != txq->qdisc_sleeping);
unsigned int i;
for (i = 0; i < dev->num_tx_queues; i++) {
struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
if (txq->qdisc != txq->qdisc_sleeping)
return true;
}
return false;
}
/* Is the device using the noop qdisc? */
/* Is the device using the noop qdisc on all queues? */
static inline bool qdisc_tx_is_noop(const struct net_device *dev)
{
const struct netdev_queue *txq = &dev->tx_queue;
return (txq->qdisc == &noop_qdisc);
unsigned int i;
for (i = 0; i < dev->num_tx_queues; i++) {
struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
if (txq->qdisc != &noop_qdisc)
return false;
}
return true;
}
static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
......
......@@ -570,16 +570,18 @@ static void vlan_dev_set_rx_mode(struct net_device *vlan_dev)
*/
static struct lock_class_key vlan_netdev_xmit_lock_key;
static void vlan_dev_set_lockdep_one(struct netdev_queue *txq,
int subclass)
static void vlan_dev_set_lockdep_one(struct net_device *dev,
struct netdev_queue *txq,
void *_subclass)
{
lockdep_set_class_and_subclass(&txq->_xmit_lock,
&vlan_netdev_xmit_lock_key, subclass);
&vlan_netdev_xmit_lock_key,
*(int *)_subclass);
}
static void vlan_dev_set_lockdep_class(struct net_device *dev, int subclass)
{
vlan_dev_set_lockdep_one(&dev->tx_queue, subclass);
netdev_for_each_tx_queue(dev, vlan_dev_set_lockdep_one, &subclass);
}
static const struct header_ops vlan_header_ops = {
......
......@@ -1666,6 +1666,12 @@ out_kfree_skb:
* --BLG
*/
static struct netdev_queue *dev_pick_tx(struct net_device *dev,
struct sk_buff *skb)
{
return netdev_get_tx_queue(dev, 0);
}
int dev_queue_xmit(struct sk_buff *skb)
{
struct net_device *dev = skb->dev;
......@@ -1702,7 +1708,7 @@ int dev_queue_xmit(struct sk_buff *skb)
}
gso:
txq = &dev->tx_queue;
txq = dev_pick_tx(dev, skb);
spin_lock_prefetch(&txq->lock);
/* Disable soft irqs for various locks below. Also
......@@ -3788,8 +3794,9 @@ static void rollback_registered(struct net_device *dev)
dev_put(dev);
}
static void __netdev_init_queue_locks_one(struct netdev_queue *dev_queue,
struct net_device *dev)
static void __netdev_init_queue_locks_one(struct net_device *dev,
struct netdev_queue *dev_queue,
void *_unused)
{
spin_lock_init(&dev_queue->_xmit_lock);
netdev_set_lockdep_class(&dev_queue->_xmit_lock, dev->type);
......@@ -3798,8 +3805,8 @@ static void __netdev_init_queue_locks_one(struct netdev_queue *dev_queue,
static void netdev_init_queue_locks(struct net_device *dev)
{
__netdev_init_queue_locks_one(&dev->tx_queue, dev);
__netdev_init_queue_locks_one(&dev->rx_queue, dev);
netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
__netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
}
/**
......@@ -4119,7 +4126,8 @@ static struct net_device_stats *internal_stats(struct net_device *dev)
}
static void netdev_init_one_queue(struct net_device *dev,
struct netdev_queue *queue)
struct netdev_queue *queue,
void *_unused)
{
spin_lock_init(&queue->lock);
queue->dev = dev;
......@@ -4127,8 +4135,8 @@ static void netdev_init_one_queue(struct net_device *dev,
static void netdev_init_queues(struct net_device *dev)
{
netdev_init_one_queue(dev, &dev->rx_queue);
netdev_init_one_queue(dev, &dev->tx_queue);
netdev_init_one_queue(dev, &dev->rx_queue, NULL);
netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
}
/**
......@@ -4145,9 +4153,10 @@ static void netdev_init_queues(struct net_device *dev)
struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
void (*setup)(struct net_device *), unsigned int queue_count)
{
void *p;
struct netdev_queue *tx;
struct net_device *dev;
int alloc_size;
void *p;
BUG_ON(strlen(name) >= sizeof(dev->name));
......@@ -4167,11 +4176,22 @@ struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
return NULL;
}
tx = kzalloc(sizeof(struct netdev_queue) * queue_count, GFP_KERNEL);
if (!tx) {
printk(KERN_ERR "alloc_netdev: Unable to allocate "
"tx qdiscs.\n");
kfree(p);
return NULL;
}
dev = (struct net_device *)
(((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
dev->padded = (char *)dev - (char *)p;
dev_net_set(dev, &init_net);
dev->_tx = tx;
dev->num_tx_queues = queue_count;
if (sizeof_priv) {
dev->priv = ((char *)dev +
((sizeof(struct net_device) +
......@@ -4205,6 +4225,8 @@ void free_netdev(struct net_device *dev)
{
release_net(dev_net(dev));
kfree(dev->_tx);
/* Compatibility with error handling in drivers */
if (dev->reg_state == NETREG_UNINITIALIZED) {
kfree((char *)dev - dev->padded);
......
......@@ -636,7 +636,7 @@ static int rtnl_fill_ifinfo(struct sk_buff *skb, struct net_device *dev,
if (dev->master)
NLA_PUT_U32(skb, IFLA_MASTER, dev->master->ifindex);
txq = &dev->tx_queue;
txq = netdev_get_tx_queue(dev, 0);
if (txq->qdisc_sleeping)
NLA_PUT_STRING(skb, IFLA_QDISC, txq->qdisc_sleeping->ops->id);
......
......@@ -621,7 +621,7 @@ int ieee80211_start_tx_ba_session(struct ieee80211_hw *hw, u8 *ra, u16 tid)
/* ensure that TX flow won't interrupt us
* until the end of the call to requeue function */
txq = &local->mdev->tx_queue;
txq = netdev_get_tx_queue(local->mdev, 0);
spin_lock_bh(&txq->lock);
/* create a new queue for this aggregation */
......@@ -862,7 +862,7 @@ void ieee80211_stop_tx_ba_cb(struct ieee80211_hw *hw, u8 *ra, u8 tid)
/* avoid ordering issues: we are the only one that can modify
* the content of the qdiscs */
txq = &local->mdev->tx_queue;
txq = netdev_get_tx_queue(local->mdev, 0);
spin_lock_bh(&txq->lock);
/* remove the queue for this aggregation */
ieee80211_ht_agg_queue_remove(local, sta, tid, 1);
......
......@@ -574,7 +574,7 @@ static struct Qdisc_ops wme_qdisc_ops __read_mostly =
void ieee80211_install_qdisc(struct net_device *dev)
{
struct netdev_queue *txq = &dev->tx_queue;
struct netdev_queue *txq = netdev_get_tx_queue(dev, 0);
struct Qdisc *qdisc;
qdisc = qdisc_create_dflt(dev, txq,
......@@ -596,7 +596,7 @@ void ieee80211_install_qdisc(struct net_device *dev)
int ieee80211_qdisc_installed(struct net_device *dev)
{
struct netdev_queue *txq = &dev->tx_queue;
struct netdev_queue *txq = netdev_get_tx_queue(dev, 0);
return txq->qdisc_sleeping->ops == &wme_qdisc_ops;
}
......@@ -617,7 +617,7 @@ int ieee80211_ht_agg_queue_add(struct ieee80211_local *local,
struct sta_info *sta, u16 tid)
{
int i;
struct netdev_queue *txq = &local->mdev->tx_queue;
struct netdev_queue *txq = netdev_get_tx_queue(local->mdev, 0);
struct ieee80211_sched_data *q =
qdisc_priv(txq->qdisc_sleeping);
DECLARE_MAC_BUF(mac);
......@@ -652,14 +652,14 @@ int ieee80211_ht_agg_queue_add(struct ieee80211_local *local,
}
/**
* the caller needs to hold local->mdev->tx_queue.lock
* the caller needs to hold netdev_get_tx_queue(local->mdev, X)->lock
*/
void ieee80211_ht_agg_queue_remove(struct ieee80211_local *local,
struct sta_info *sta, u16 tid,
u8 requeue)
{
struct ieee80211_hw *hw = &local->hw;
struct netdev_queue *txq = &local->mdev->tx_queue;
struct netdev_queue *txq = netdev_get_tx_queue(local->mdev, 0);
struct ieee80211_sched_data *q =
qdisc_priv(txq->qdisc_sleeping);
int agg_queue = sta->tid_to_tx_q[tid];
......@@ -676,7 +676,7 @@ void ieee80211_ht_agg_queue_remove(struct ieee80211_local *local,
void ieee80211_requeue(struct ieee80211_local *local, int queue)
{
struct netdev_queue *txq = &local->mdev->tx_queue;
struct netdev_queue *txq = netdev_get_tx_queue(local->mdev, 0);
struct Qdisc *root_qd = txq->qdisc_sleeping;
struct ieee80211_sched_data *q = qdisc_priv(root_qd);
struct Qdisc *qdisc = q->queues[queue];
......
......@@ -74,14 +74,16 @@ static const struct proto_ops nr_proto_ops;
*/
static struct lock_class_key nr_netdev_xmit_lock_key;
static void nr_set_lockdep_one(struct netdev_queue *txq)
static void nr_set_lockdep_one(struct net_device *dev,
struct netdev_queue *txq,
void *_unused)
{
lockdep_set_class(&txq->_xmit_lock, &nr_netdev_xmit_lock_key);
}
static void nr_set_lockdep_key(struct net_device *dev)
{
nr_set_lockdep_one(&dev->tx_queue);
netdev_for_each_tx_queue(dev, nr_set_lockdep_one, NULL);
}
/*
......
......@@ -75,14 +75,16 @@ ax25_address rose_callsign;
*/
static struct lock_class_key rose_netdev_xmit_lock_key;
static void rose_set_lockdep_one(struct netdev_queue *txq)
static void rose_set_lockdep_one(struct net_device *dev,
struct netdev_queue *txq,
void *_unused)
{
lockdep_set_class(&txq->_xmit_lock, &rose_netdev_xmit_lock_key);
}
static void rose_set_lockdep_key(struct net_device *dev)
{
rose_set_lockdep_one(&dev->tx_queue);
netdev_for_each_tx_queue(dev, rose_set_lockdep_one, NULL);
}
/*
......
......@@ -166,7 +166,7 @@ replay:
/* Find qdisc */
if (!parent) {
struct netdev_queue *dev_queue = &dev->tx_queue;
struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, 0);
q = dev_queue->qdisc_sleeping;
parent = q->handle;
} else {
......@@ -410,7 +410,7 @@ static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
return skb->len;
dev_queue = &dev->tx_queue;
dev_queue = netdev_get_tx_queue(dev, 0);
if (!tcm->tcm_parent)
q = dev_queue->qdisc_sleeping;
else
......
......@@ -183,9 +183,8 @@ EXPORT_SYMBOL(unregister_qdisc);
(root qdisc, all its children, children of children etc.)
*/
struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
static struct Qdisc *__qdisc_lookup(struct netdev_queue *dev_queue, u32 handle)
{
struct netdev_queue *dev_queue = &dev->tx_queue;
struct Qdisc *q;
list_for_each_entry(q, &dev_queue->qdisc_list, list) {
......@@ -195,6 +194,19 @@ struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
return NULL;
}
struct Qdisc *